Demo of RNeo4j Part 2: Plotting and Analysis

In Part 1, I built a database using RNeo4j and twitteR. Now I can use RNeo4j to retreive data from the database for plotting and analysis using a few of R’s many available packages.

Hashtag Co-Occurrence with #Neo4j

My favorite function in RNeo4j, cypher, allows you to retrieve Cypher query results as an R data frame. The following query gets the 10 hashtags that co-occur with #neo4j most frequently and their counts into a data frame called hashtag_count.

query = "MATCH (h:Hashtag)<-[:HASHTAG]-(:Tweet)-[:HASHTAG]->(:Hashtag {hashtag:{hashtag}}) 
         WHERE h.hashtag <> {hashtag}
         RETURN h.hashtag AS hashtag, COUNT(*) AS count
         ORDER BY count DESC
         LIMIT 10"

hashtag_count = cypher(graph, query, hashtag = "#neo4j")

hashtag_count

#           hashtag count
# 1         #spring    10
# 2      #graphunit     9
# 3         #vaadin     8
# 4         #munich     7
# 5  #graphdatabase     7
# 6       #graphpub     6
# 7        #graphdb     6
# 8   #graphhackday     6
# 9     #graphaware     6
# 10         #graph     5

With the results in a data frame, I can easily create charts using ggplot2. If I wanted to plot the information in hashtag_count in a bar chart, for example, I can do so with the following.

install.packages("ggplot2")
library(ggplot2)

ggplot(hashtag_count, aes(x = reorder(hashtag, count), y = count)) + 
  geom_bar(stat = "identity", fill = "darkblue") +
  coord_flip() +
  labs(x = "Hashtag", 
       y = "Count", 
       title = "Count of Hashtag Co-Occurrence with #Neo4j") +
  theme(axis.text = element_text(size = 12, color = "black"),
        axis.title = element_text(size = 14, color = "black"),
        plot.title = element_text(size = 16, color = "black"))

Word Cloud

I decided it would be fun to create a word cloud of all the tweets’ text using the wordcloud, tm, and RColorBrewer packages.

First, install and load the necessary packages.

install.packages(c("wordcloud", "tm", "RColorBrewer"))

library(wordcloud)
library(tm)
library(RColorBrewer)

Next I need to get all of the Tweet nodes from the graph into a list of node objects. This can be done with RNeo4j’s getLabeledNodes function, which gets all nodes with the specified label.

tweet_nodes = getLabeledNodes(graph, "Tweet")

Each node object’s properties can be accessed by node$property. For example, the text of the first node object in tweet_nodes is accessed through the following.

tweet_nodes[[1]]$text

# "The beauty of Tom Sawyer Perspectives is that it can not only pull data from the Neo4j database, http://t.co/FA46LO5U6k #DataViz"

Use sapply to extract the text property from each node object in tweet_nodes.

tweet_text = sapply(tweet_nodes, function(t) t$text)

tweet_text is a character vector of all the tweets’ text. Now I switch over to the tm package to remove stopwords, punctuation, etc., and then to the RColorBrewer and wordcloud packages to create the word cloud. Most of this code is shamelessly copied from the documentation for the wordcloud function.

# Remove links and convert to lowercase.
tweet_text = sapply(tweet_text, function(t) str_replace_all(t, perl("http.+?(?=(\\s|$))"), ""))
tweet_text = tolower(tweet_text)

# Remove stopwords, punctuation, etc.
tweet_corpus = Corpus(VectorSource(tweet_text))
tweet_corpus = tm_map(tweet_corpus, removePunctuation)
tweet_corpus = tm_map(tweet_corpus, function(x) removeWords(x, c(stopwords("english"), "neo4j")))

# Get term-document matrix and then a term-frequency data frame.
tdm = TermDocumentMatrix(tweet_corpus)
m = as.matrix(tdm)
v = sort(rowSums(m),decreasing=TRUE)
d = data.frame(word = names(v),freq=v)

# Create the word cloud.
pal = brewer.pal(9,"Dark2")
wordcloud(words = d$word, 
          freq = d$freq, 
          scale = c(8,.3), 
          random.order = F,
          colors = pal)